Recent Advances, Emerging Methods and Applications of Pattern Recognition
J.UCS Special Issue
Jörg Keller
(Faculty of Mathematics and Computer Science, FernUniversität in Hagen, Germany
joerg.keller@fernuni-hagen.de)
Manuel Grana
(Faculty of Informatics, UPV/EHU University of the Basque Country, Donostia-San Sebastian, Spain
Manuel.grana@ehu.es)
Rafał Kozik
(University of Science and Technology, UTP Bydgoszcz, Poland
rkozik@utp.edu.pl)
Tomasz Andrysiak
(University of Science and Technology, UTP Bydgoszcz, Poland
andrys@utp.edu.pl)
Michał Choraś
(University of Science and Technology, UTP Bydgoszcz, Poland
chorasm@utp.edu.pl)
Nowadays advanced pattern extraction and analysis techniques have been
successfully used in a variety of domains. Nonetheless, many
challenges still exist. In particular, the majority of practical
applications of pattern recognition are still struggling with huge
amounts of data that exhibits such challenges as concept drift,
non-stationary underlying processes, noise, and common lack of labels
or additional metadata that could be useful for further analysis.
In addition to that, scalability of those practical applications and
ability to adapt to constantly changing environment is still a
problem. Moreover, problems with heterogeneous and multi-source data
that require a dedicated approach also emerge.
Therefore, in this Special Issue, we covered recent solutions to the
abovementioned problems, novel methods and algorithms, advances,
challenges, as well as practical applications of pattern recognition.
The special issue includes, among other submissions from the open call
for papers, extended versions of accepted papers from the 10th
International Conference on Image Conference and Communications -
IP&C 2018. IP&C is a known Image Processing and Communications
conference organized at the University of Science and Technology, UTP
Bydgoszcz, Poland since 2009. Those extended papers have been invited
for submission under the condition of providing at least 50% new
content. All 14 submissions were peer-reviewed by top experts in the
domain. After revision and re-review, based on the reviews and the
judgement of the guest editors, 9 articles were selected for
publication in this special issue to represent the breadth of the
field.
Łukasz Apiecionek, Marcel Großmann and Udo R. Krieger presented IoT
architectures and compared them focusing on security aspects.
Sławomir Bujnowski, Tomasz Marciniak, Beata Marciniak and Zbignew
Lutowski analysed the influence of the network resources control on
the transmission properties of the networks. They particularly focused
on topologies that are described with irregular graphs.
Dominik Pieczyński, Marek Kraft and Michał Fularz focused on improving
the quality of person re-identification with proposed deep learning
tools. They hypothesised that including segmentation information in
the processing pipeline allows for discarding poor quality detections.
Hubert Michalak and Krzysztof Okarma advance the state of the art of
optical character recognition in documents with uneven illumination by
means of an innovative local thresholding algorithm of proven
robustness.
Waseem Rawat and Zenghui Wang tackle the difficult problem of topology
design in the case of convolutional neural networks applying a
bayesian search embedded in a genetic algorithm approach with improved
results in a well known benchmarking dataset.
Sonia Contreras, Miguel Ángel Manzenado and Álvaro Herrero provide a
highly valuable application of hybrid neural system from the social
and economical point view, that of analyzing the relation between
workplace accidents and the global economical crisis scenario.
Michał Sypetkowski, Grzegorz Sarwas and Tomasz Trzciński present the
problem of football players poses recorded and presented on low -
resolution images, received from HQ CCTV system located on lighting
spots, which are considered to be very efficient and easy to operate.
Jaroslaw Fastowicz and Krzysztof Okarma analyze different techniques
of 3D modelling and printing, which results seem to be very
promising. Among others, they draw attention to the advantage of low
dependence on the color of the filament for the popular PLA and ABS
materials, used typically in different (also: of low budget) devices.
Robert Burduk and Jedrzej Biedrzycki touch upon the case of
introducing more sophisticated and developed algorithm of classifier
integration in geometric space. In the suggested solution, decision
boundaries in the process of integrating instead of class labels or
predicted probabilities may help to achieve better results.
We would like to express our thankfulness to Christian Guütl (Managing
Editor) and Dana Kaiser (Head of Editorial Team) for permitting us to
organize this special issue under the umbrella of the Journal of
Universal Computer Science.
We also like to thank all reviewers who facilitated the review
process, namely: Darius Andriukaitis, Darya Chyzhyk, Adam Czajka,
Panayiotis Frangos, Petra Grd, Agata Giełczyk, A.I. Gonzalez, Jose
Gutierrez, Sebastian Litzinger, Javier De Lope, Reza Malekian,
Wojciech Mazurczyk, Marek Pawlicki, Jens Myrup Pedersen, Maciej
Piechowiak, Bogdan Raducanu, Adam Schmidt.
|